Stef van Buuren

Multiple imputation of discrete and continuous data by fully conditional specification

The goal of multiple imputation is to provide valid inferences for statistical estimates from incomplete data. To achieve that goal, imputed values should preserve the structure in the data, as well as the uncertainty about this structure, and include any knowledge about the process that generated the missing data. Two approaches for imputing multivariate data exist: joint modeling (JM) and fully conditional specification (FCS). JM is based on parametric statistical theory, and leads to imputation procedures whose statistical properties are known. JM is theoretically sound, but the joint model may lack flexibility needed to represent typical data features, potentially leading to bias. FCS is a semi-parametric and flexible alternative that specifies the multivariate model by a series of conditional models, one for each incomplete variable. FCS provides tremendous flexibility and is easy to apply, but its statistical properties are difficult to establish. Simulation work shows that FCS behaves very well in ...The goal of multiple imputation is to provide valid inferences for statistical estimates from incomplete data. To achieve that goal, imputed values should preserve the structure in the data, as well as the uncertainty about this structure, and include any knowledge about the process that generated the missing data. Two approaches for imputing multivariate data exist: joint modeling (JM) and fully conditional specification (FCS). JM is based on parametric statistical theory, and leads to imputation procedures whose statistical properties are known. JM is theoretically sound, but the joint model may lack flexibility needed to represent typical data features, potentially leading to bias. FCS is a semi-parametric and flexible alternative that specifies the multivariate model by a series of conditional models, one for each incomplete variable. FCS provides tremendous flexibility and is easy to apply, but its statistical properties are difficult to establish. Simulation work shows that FCS behaves very well in the cases studied. The present paper reviews and compares the approaches. JM and FCS were applied to pubertal development data of 3801 Dutch girls that had missing data on menarche (two categories), breast development (five categories) and pubic hair development (six stages). Imputations for these data were created under two models: a multivariate normal model with rounding and a conditionally specified discrete model. The JM approach introduced biases in the reference curves, whereas FCS did not. The paper concludes that FCS is a useful and easily applied flexible alternative to JM when no convenient and realistic joint distribution can be specified.

Are age references for waist circumference, hip circumference and waist-hip ratio in Dutch children useful in clinical practice?

The aim of this study was to present age references for waist circumference (WC), hip circumference (HC), and waist/hip ratio (WHR) in Dutch children. Cross-sectional data were obtained from 14,500 children of Dutch origin in the age range 0–21 years. National references were constructed with the LMS method. This method summarises the distribution by three smooth curves representing skewness (L curve), the median (M curve), and coefficient of variation (S curve). The correlations between body mass index-standard deviation score (BMI-SDS), the circumferences and their ratio, and demographic variables were assessed by (multiple) regression analysis for three age groups: 0-<5 years (1), 5-<12.5 years (2), and 12.5-<21 years (3). A cut-off for clinical use was suggested based on the International Obesity Task Force criteria for BMI. Mean WC and HC values increased with age. Mean WC was slightly higher in boys than in girls, and this difference was statistically significant from 11 years of age onwards. In contrast, HC was significantly higher in girls than in boys from 9 years onwards. The correlation between WC-SDS and BMI-SDS ( r =0.73, P <0.01) and between HC and BMI-SDS ( r =0.67, P <0.01) increased with age. With regard to WHR-SDS, a low correlation was found for 12.5–20 years of age ( r =0.2, P <0.01). WC-SDS correlated positively with height SDS ( r =0.35, P <0.01). Conclusion:Waist circumferences can be used to screen for increased abdominal fat mass in children, whereby a cut-off point of 1.3 standard deviation score seems most suitable.

Pubertal development in The Netherlands 1965-1997.

We investigated pubertal development of 4019 boys and 3562 girls >8 y of age participating in a cross-sectional survey in The Netherlands and compared the results with those of two previous surveys. Reference curves for all pubertal stages were constructed. The 50th percentile of Tanner breast stage 2 was 10.7 y, and 50% of the boys had reached a testicular volume of 4 mL at 11.5 y of age. Median age at menarche was 13.15 y. The median age at which the various stages of pubertal development were observed has stabilized since 1980. The increase of the age at stage G2 between 1965 and 1997 is probably owing to different interpretations of its definition. The current age limits for the definition of precocious are close to the third percentile of these references. A high agreement was found between the pubic hair stages and stages of pubertal (genital and breast) development, but slightly more in boys than in girls. Menarcheal age was dependent on height, weight, and body mass index. At a given age tall or heavy girls have a higher probability of having menarche compared with short or thin girls. A body weight exceeding 60 kg (+1 SDS), or a body mass index of >20 (+1 SDS), has no or little effect on the chance of having menarche, whereas for height such a ceiling effect was not observed. In conclusion, in The Netherlands the age at onset of puberty or menarche has stabilized since 1980. Height, weight, and body mass index have a strong influence on the chance of menarche.

Increase in Prevalence of Overweight in Dutch Children and Adolescents: A Comparison of Nationwide Growth Studies in 1980, 1997 and 2009

Objective To assess the prevalence of overweight and obesity among Dutch children and adolescents, to examine the 30-years trend, and to create new body mass index reference charts. Design Nationwide cross-sectional data collection by trained health care professionals. Participants: 10,129 children of Dutch origin aged 0–21 years. Main Outcome Measures Overweight (including obesity) and obesity prevalences for Dutch children, defined by the cut-off values on body mass index references according to the International Obesity Task Force. Results In 2009, 12.8% of the Dutch boys and 14.8% of the Dutch girls aged 2–21 years were overweight and 1.8% of the boys and 2.2% of the girls were classified as obese. This is a two to three fold higher prevalence in overweight and four to six fold increase in obesity since 1980. Since 1997, a substantial rise took place, especially in obesity, which increased 1.4 times in girls and doubled in boys. There was no increase in mean BMI SDS in the major cities since 1997. Conclusions Overweight and obesity prevalences in 2009 were substantially higher than in 1980 and 1997. However, the overweight prevalence stabilized in the major cities. This might be an indication that the rising trend in overweight in the Netherlands is starting to turn.

Reference chart for relative weight change to detect hypernatraemic dehydration

Objective: The validity of the rule of thumb that infants may have a weight loss of 10% in the first days after birth is unknown. We assessed the validity of this and other rules to detect breast-fed infants with hypernatraemic dehydration. Design: A reference chart for relative weight change was constructed by the LMS method. The reference group was obtained by a retrospective cohort study. Participants: 1544 healthy, exclusively breast-fed infants with 3075 weight measurements born in the Netherlands and 83 cases of breast-fed infants with hypernatraemic dehydration obtained from literature. Results: The rule of thumb had a sensitivity of 90.4%, a specificity of 98.3% and a positive predictive value of 3.7%. Referring infants if their weight change is below −2.5 SDS (0.6th centile) in the reference chart in the first week of life and using the rule of thumb in the second week had a sensitivity of 85.5%, a specificity of 99.4% and a positive predictive value of 9.2%. Conclusions: The rule of thumb is likely to produce too many false positive results, assuming that for screening purposes the specificity needs to be high. A chart for relative weight change can be helpful to detect infants with hypernatraemic dehydration.

Alarming prevalences of overweight and obesity for children of Turkish, Moroccan and Dutch origin in The Netherlands according to international standards.

Aim: Prevalences of overweight in The Netherlands, defined by international cut‐off points, are presented in 14 500 children of Dutch origin, 2904 of Turkish and 2855 of Moroccan origin, aged 0–21 y. Results: The mean prevalence for Turkish boys and girls was 23.4% and 30.2%, for Moroccans 15.8% and 24.5%, for Dutch youths in large cities 12.6% and 16.5%, and for other Dutch participants 8.7% and 11.3%, respectively.

Prevalence of overweight and obesity in the Netherlands in 2003 compared to 1980 and 1997

Objective: To assess the prevalence of overweight and obesity in children living in the Netherlands and compare the findings with the Third and Fourth National Growth Studies carried out in 1980 and 1997, respectively. Design and methods: Data were obtained from the child health care system. International cut-off points for body mass index (BMI) were used to determine overweight and obesity. Cases were weighted for ethnicity and municipality size in such a way that the sample matched the distribution in the general population. The LMS method was used to calculate the age-related distribution of BMI, and the prevalence was calculated from the fitted distribution. Patients: Data on 90 071 children aged 4–16 years were routinely collected by 11 community health services during 2002–2004. Results: On average, 14.5% of the boys and 17.5% of the girls were overweight (including obesity), which is a substantial increase since 1980 (boys 3.9%, girls 6.9%) and 1997 (boys 9.7%, girls 13.0%). Similarly, 2.6% of the boys and 3.3% of the girls aged 4–16 years were obese, which is much higher than in 1980 (boys 0.2%, girls 0.5%) and 1997 (boys 1.2%, girls 2.0%). At the age of 4, 12.3% of the boys and 16.2% of the girls were already overweight. Conclusions: The prevalence of overweight and obesity in the Netherlands is still rising, and at an even faster rate than before. Evidence-based interventions are needed to counter the obesity epidemic, and there is an urgent need for pre-school intervention programmes.

Association between parenting practices and children's dietary intake, activity behavior and development of body mass index: the KOALA Birth Cohort Study.

BackgroundInsights into the effects of energy balance-related parenting practices on childrens diet and activity behavior at an early age is warranted to determine which practices should be recommended and to whom. The purpose of this study was to examine child and parent background correlates of energy balance-related parenting practices at age 5, as well as the associations of these practices with childrens diet, activity behavior, and body mass index (BMI) development.MethodsQuestionnaire data originated from the KOALA Birth Cohort Study for ages 5 (N = 2026) and 7 (N = 1819). Linear regression analyses were used to examine the association of child and parent background characteristics with parenting practices (i.e., diet- and activity-related restriction, monitoring and stimulation), and to examine the associations between these parenting practices and childrens diet (in terms of energy intake, dietary fiber intake, and added sugar intake) and activity behavior (i.e., physical activity and sedentary time) at age 5, as well as BMI development from age 5 to age 7. Moderation analyses were used to examine whether the associations between the parenting practices and child behavior depended on child characteristics.ResultsSeveral child and parent background characteristics were associated with the parenting practices. Dietary monitoring, stimulation of healthy intake and stimulation of physical activity were associated with desirable energy balance-related behaviors (i.e., dietary intake and/or activity behavior) and desirable BMI development, whereas restriction of sedentary time showed associations with undesirable behaviors and BMI development. Child eating style and weight status, but not child gender or activity style, moderated the associations between parenting practices and behavior. Dietary restriction and monitoring showed weaker, or even undesirable associations for children with a deviant eating style, whereas these practices showed associations with desirable behavior for normal eaters. By contrast, stimulation to eat healthy worked particularly well for children with a deviant eating style or a high BMI.ConclusionAlthough most energy balance-related parenting practices were associated with desirable behaviors, some practices showed associations with undesirable child behavior and weight outcomes. Only parental stimulation showed desirable associations with regard to both diet and activity behavior. The interaction between parenting and child characteristics in the association with behavior calls for parenting that is tailored to the individual child.

Variable selection under multiple imputation using the bootstrap in a prognostic study

BackgroundMissing data is a challenging problem in many prognostic studies. Multiple imputation (MI) accounts for imputation uncertainty that allows for adequate statistical testing. We developed and tested a methodology combining MI with bootstrapping techniques for studying prognostic variable selection.MethodIn our prospective cohort study we merged data from three different randomized controlled trials (RCTs) to assess prognostic variables for chronicity of low back pain. Among the outcome and prognostic variables data were missing in the range of 0 and 48.1%. We used four methods to investigate the influence of respectively sampling and imputation variation: MI only, bootstrap only, and two methods that combine MI and bootstrapping. Variables were selected based on the inclusion frequency of each prognostic variable, i.e. the proportion of times that the variable appeared in the model. The discriminative and calibrative abilities of prognostic models developed by the four methods were assessed at different inclusion levels.ResultsWe found that the effect of imputation variation on the inclusion frequency was larger than the effect of sampling variation. When MI and bootstrapping were combined at the range of 0% (full model) to 90% of variable selection, bootstrap corrected c-index values of 0.70 to 0.71 and slope values of 0.64 to 0.86 were found.ConclusionWe recommend to account for both imputation and sampling variation in sets of missing data. The new procedure of combining MI with bootstrapping for variable selection, results in multivariable prognostic models with good performance and is therefore attractive to apply on data sets with missing values.

Trends in Menarcheal Age between 1955 and 2009 in the Netherlands

Aim To assess and compare the secular trend in age at menarche in Dutch girls (1955–2009) and girls from Turkish and Moroccan descent living in the Netherlands (1997–2009). Methods Data on growth and maturation were collected in 20,867 children of Dutch, Turkish and Moroccan descent in 2009 by trained health care professionals. Girls, 9 years and older, of Dutch (n = 2138), Turkish (n = 282), and Moroccan (n = 295) descent were asked whether they had experienced their first period. We compared median menarcheal age in 2009 with data from the previous Dutch Nationwide Growth Studies in 1955, 1965, 1980 and 1997. Age specific body mass index (BMI) z-scores were calculated to assess differences in BMI between pre- and postmenarcheal girls in different age groups. Results Median age at menarche in Dutch girls, decreased significantly from 13.66 years in 1955 to 13.15 years in 1997 and 13.05 years in 2009. Compared to Dutch girls there is a larger decrease in median age of menarche in girls of Turkish and Moroccan descent between 1997 and 2009. In Turkish girls age at menarche decreased from 12.80 to 12.50 years and in Moroccan girls from 12.90 to 12.60 years. Thirty-three percent of Turkish girls younger than 12 years start menstruating in primary school. BMI-SDS is significantly higher in postmenarcheal girls than in premenarcheal girls irrespective of age. Conclusion There is a continuing secular trend in earlier age at menarche in Dutch girls. An even faster decrease in age at menarche is observed in girls of Turkish and Moroccan descent in the Netherlands.